Atomic Structure of Luminescent Centers in High-Efficiency Ce-doped w-AlN Single Crystal

Rare-earth doped wurtzite-type aluminum nitride (w-AlN) has great potential for high-efficiency electroluminescent applications over a wide wavelength range. However, because of their large atomic size, it has been difficult to stably dope individual rare-earth atoms into the w-AlN host lattice. Here we use a reactive flux method under high pressure and high temperature to obtain cerium (Ce) doped w-AlN single crystals with pink-colored luminescence. In order to elucidate the atomic structure of the luminescent centers, we directly observe individual Ce dopants in w-AlN using annular dark-field scanning transmission electron microscopy. We find that Ce is incorporated as single, isolated atoms inside the w-AlN lattice occupying Al substitutional sites. This new synthesis method represents a new alternative strategy for doping size-mismatched functional atoms into wide band-gap materials.

The present w-AlN:Ce single crystals were synthesized by the temperature gradient method at 6.5 GPa and 1600uC. The resulting sub-millimeter w-AlN:Ce single crystals (,0.2 mm cube) show optical transparency and the band edge luminescent feature of w-AlN is observed at 210 nm in cathodoluminescence (CL) spectroscopy, suggesting that the impurity level is sufficiently suppressed. Figure 1(a) shows a photoluminescence (PL) image of w-AlN:Ce single crystals. The homogeneous pink-colored luminescence shows that the Ce dopants are uniformly doped throughout the single crystals. A single broadband peak, observed at 600 nm in the PL spectrum of Figure 1 (b), is assigned to the 4f -5d electron transition of Ce 31 dopants in a bulk material. The valence state of Ce dopants in the w-AlN single crystals was evaluated by Ce M 4,5 X-ray absorption near edge structure (XANES). Two specific peaks in Figure 1(c) are assigned to the Ce M 5 (879 eV) and M 4 (897 eV) edges. By comparing to the standard XANES spectra of Ce 31 in CeF 3 and Ce 41 in CeO 2 , the valence state of the Ce dopants in w-AlN is determined to be 31. Figure 2(a) shows a typical atomic-resolution ADF STEM image of w-AlN:Ce viewed along the 11 20 ½ direction. The scattering strength of nitrogen is relatively weak, but the atomic dumbbell consisting of Al and N atoms is clearly resolved (the projected spacing is 1.1 Å ), as is evident in Figure 2(b). Thus, we can directly determine the atomic site of the Ce dopants from the images, whether they are on Al, N or interstitial sites. As marked by the arrows in Figure 2(a), the atomic columns containing Ce dopants are observed as slightly brighter spots, and one can see that the Ce dopants are dispersed as single atoms. This isolation of single Ce dopants is critical for the optical luminescence because dopant-clusters lose radiative recombination efficiency, in particular for large clusters, owing to their local metallic properties. The present rare-earth doping process in w-AlN is thus highly effective, and all Ce dopants should contribute equally to the pink-colored luminescence. Note that, even after the extensive observations, we could not observe any rare-earth dimer-or trimer-clusters, as previously proposed 12,14 , suggesting that these clusters may be meta-stable defect structures. In the high-magnification atomicresolution ADF STEM image of Figure 2(b), the single Ce dopant evidently substitutes for the Al site without apparent atomic displacements. This bright feature at the Al site is quantitatively reproduced in the simulated image ( Figure 2(c)), where a single Ce atom has been substituted for an Al atomic site.
In order to theoretically estimate the stable Ce atom configurations in the w-AlN lattice, we performed systematic first-principles calculations using the Heyd-Scuseria-Ernzerhof (HSE06) hybrid functional 20,21 , which has been applied to Ce dopants in c-BN 18 as well as a variety of solids including rare-earth compounds [22][23][24] . The formation energy (E f ) of a defect is written as where E def T denotes the total energy of the supercell (192 atoms) with a defect in charge state q, n i is the number of the constituent atoms of i-type, and m i and E F are the atomic chemical potential and the Fermi level. The chemical potentials for the w-AlN:Ce single crystals are not clear under the present synthesis conditions, and therefore we treated the chemical potentials as variables. m Al , m Ce , and m N are considered to range between the following extreme conditions: the Al-rich limit (m Al 5 m Al(bulk) , m N 5 1/2m N2 1 DE f (AlN), and m Ce 5 m Ce(bulk) ) and the N-rich limit (m N where Ce Al and V N are located along the direction parallel to or nearly perpendicular to the c axis). In addition, isolated Al and N vacancies (V Al and V N ) were considered as the dominant native defects in w-AlN. Figures 3(a) and (b) show the formation energies of several defect structures and complexes considered here as a function of the Fermi level. The formation energies of V N and V Al (dashed lines in Figures 3(a) and (b)) are negative in the lower Fermi levels (, 1.2 eV) at the Al-rich limit and in the higher Fermi levels (. 4.4 eV) at the N-rich limit, respectively. The Fermi level cannot be located in these ranges because of carrier compensation associated with the spontaneous formation of negatively charged V Al and positively charged V N . The XANES measurement suggests the presence of Ce 31 in the w-AlN:Ce single crystals. From the formation energy and one-electron structure analyses, the energetically stable defect structures that involve Ce 31 are (1)   Á 2{ defect complex also shows a lower formation energy than Ce Al , but the condition is limited to very high Fermi levels (. 5.2 eV) and only found at the Al-rich limit. In the relaxed Ce Al {V \ N À Á 2{ defect structure, the Ce atom exhibits a substantial atomic shift toward the N vacancy site from the Al site (0.23 Å ). On the other hand, neutral Ce Al is the most energetically stable defect structure under the remaining wide range of conditions (the chemical potential only slightly affects the formation energy). As seen in the relaxed structure (Figures 3(c) and (d)), the Ce atom is located at the Al site without apparent atomic displacement, which is in good agreement with the observed ADF STEM images. Thus, we conclude that the atomic structure luminescent center in w-AlN is neutral Ce Al . It is noted that the Al-N bond length (1.89 Å ) is ,29% shorter than that of Ce-N (2.44 Å ) in the compound CeN. Hence, it may appear to be impossible to accommodate an isolated Ce atom into the w-AlN lattice. However, in the neutral Ce Al defect structure, the tetrahedrally coordinated N atoms shift their positions and the resultant Ce-N bond length (2.22 Å ) is ,17% longer than that of Al-N. We note here that the theoretical formation energy of a neutral Ce Al defect is comparatively high (,3 eV) and so the doping of Ce atoms into the w-AlN lattice would seem to be very unlikely under ambient conditions. However, the present high pressure and high temperature extreme conditions are able to overcome the high formation energy of these Ce Al luminescent centers.
In summary, we have synthesized Ce-doped w-AlN single crystals with high efficiency pink-colored luminescent centers through an optimized high pressure, high temperature flux method. The atomic structure of the luminescent center in w-AlN has been determined through direct observation of atomic-resolution ADF STEM imaging combined with systematic first-principles calculations. Our findings indicate that the present synthesis method enables the stable doping of isolated single Ce atoms and could be extended to control the emission color by choosing appropriate dopant elements. Moreover, it opens a new alternative strategy for doping size-mismatched functional atoms into wide band-gap materials, overcoming the high formation energies.

Methods
Single crystal synthesis. The single crystals of w-AlN:Ce were synthesized by the temperature gradient crystal growth method at 6.5 GPa and 1600uC for 20 hours, using a modified belt-type high-pressure apparatus with a bore diameter of 30 mm (Ref. 16,17). The source of AlN powder (Toyo Soda, Type F) placed at upper part in the sample chamber was dissolved into the molten solvent and w-AlN crystals were precipitated at the cooler bottom in the chamber via a spontaneous nucleation process. We used a mixture of Li 3 AlN 2 and Ba 3 Al 2 N 4 as a solvent (typically, 151 molar ratio). In order to suppress carbon and oxygen contamination, we prepared solvents at 900uC under dry nitrogen atmosphere, and then high-purity CeF 3 powder (Rare metallic Co. Ltd, 4N grade, 0.5 wt%) as a dopant source was also mixed under the same atmosphere. The fundamental procedure for the crystal growth is similar to that of cubic boron nitride (c-BN) single crystals, where the solvents are Li 3 BN 2 and Ba 3 B 2 N 4 (Ref. 16,17). The yields of w-AlN crystals are essentially 100% because the source of AlN powder was entirely dissolved into the solvents and precipitated as recrystallized w-AlN during growth duration, though the size of the crystals depend upon the position in the Mo sample chamber. Typical weights of the source and the solvent are 0.15 g and 0.18 g.
Electron microscope experiments and simulations. The surface of the obtained single crystals was cleaned by hot aqua regia. To directly image Ce dopants in the w-AlN lattice, an important prerequisite is to prepare clean and electron-transparent thin TEM specimens (,10 nm). To avoid any surface damage, we did not use conventional Ar-ion thinning but gently crushed the single crystals in ethanol and dispersed them onto an amorphous carbon grid. To remove hydrocarbon contamination, the grid was baked at 160uC for 8 hours in a clean vacuum and then transferred into a microscope. The atomic-resolution ADF STEM images were acquired with an aberration corrected Nion UltraSTEM 200, operated at 200 kV. To avoid beam damage, atomic-resolution observations were performed under a low beam current condition (,9 pA). Under this condition, no significant change can be recognized even scanning more than 30 times over the same region. The image simulation was carried out using the frozen phonon model with a probe-forming aperture half-angle of 30 mrad, the half-angle of the ADF detector spanning 63 to 409 mrad, and incorporating the experimentally-measured, non-uniform detector efficiency 25 .
XANES experiments. The XANES measurements were carried out using the total electron yield method at BL11A in KEK-PF, Tsukuba, Japan.
First-principles calculations. The calculations were performed using the projector augmented-wave method 26 and the HSE06 hybrid functional 20,21 as implemented in the VASP code 27,28 . The defects were modeled using 192-atom supercells and spin polarization was allowed for all the defect species and charge states. A plane-wave cutoff energy of 400 eV and the C-point only k-point sampling were used in the calculations. To correct finite-size effects of the supercells with charged defects, approximate third-order image charge corrections reported in Ref. 29 were applied in conjunction with electrostatic potential alignment suggested in Ref. 30.